Inception has launched Mercury 3, a large language model built on diffusion rather than generating text one token at a time. The company describes it as its most capable model yet, claiming 1,068 tokens per second on NVIDIA GPUs and a 1 million token context window for working with very large documents and datasets.
Diffusion, mostly used so far for image generation, produces text in several passes: the model drafts the whole answer and refines it step by step. The claimed advantage is speed, which matters for voice bots and agents where latency decides whether a product feels usable.
Benchmarks and comparisons
Inception says Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 on benchmarks while running four to eight times faster. According to the company, it beats both rivals on GPQA Diamond, IFBench and SWE-Bench Atlas. These are the company's own claims and have not been independently verified. The announcement also states that Mercury 3 is more than 40 times larger than its predecessor, Mercury 1, while keeping comparable speeds.
Pricing and availability
The standard price is $0.20 per million input tokens and $0.70 per million output tokens. A 50% launch discount brings introductory rates to $0.10 and $0.35. The model is available through Inception's API, OpenRouter, Baseten and Vercel, and it is compatible with the OpenAI API protocol, so teams already using that standard can migrate without rewriting their code.
For decentralized finance and smart contract work, the model is pitched as a tool for audits and large-scale data synthesis. Whether the claimed performance holds up in real use remains to be seen.
No comments yet. Be the first!